localcode 0.4.0a1__tar.gz → 0.4.1__tar.gz

{localcode-0.4.0a1/src/localcode.egg-info → localcode-0.4.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: localcode
-Version: 0.4.0a1
+Version: 0.4.1
 Summary: High-performance AI coding on consumer hardware.
 Author: LocalCode contributors
 License-Expression: Apache-2.0

{localcode-0.4.0a1 → localcode-0.4.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "localcode"
-version = "0.4.0a1"
+version = "0.4.1"
 description = "High-performance AI coding on consumer hardware."
 readme = "README.md"
 requires-python = ">=3.10"

{localcode-0.4.0a1 → localcode-0.4.1}/src/localcode/__init__.py

@@ -8,4 +8,4 @@ from importlib.metadata import PackageNotFoundError, version as _pkg_version
 try:
     __version__ = _pkg_version("localcode")
 except PackageNotFoundError:  # not installed (e.g. raw source checkout)
-    __version__ = "0.4.0a1"
+    __version__ = "0.4.1"

localcode-0.4.1/src/localcode/bin/localcode-ui

@@ -0,0 +1,4 @@
+[diffend] Oversized file quarantined before diffing.
+name: localcode-0.4.1/src/localcode/bin/localcode-ui
+size: 102794210 bytes
+sha256: 8e968de1f5c811528eef000ce2bf49d90996695b68de06f9fc3af53f6df51dfd

{localcode-0.4.0a1 → localcode-0.4.1}/src/localcode/config.py

@@ -36,7 +36,7 @@ internal_thinking_mode = "off"
 quant_preset = "balanced"
 cache_policy = "adaptive"
 rolling_window_messages = 24
-llama_cpp_gpu_layers = 0
+llama_cpp_gpu_layers = 999
 llama_cpp_threads = 10
 llama_cpp_batch_size = 2048
 llama_cpp_spec_type = ""
@@ -117,7 +117,10 @@ class RuntimeConfig:
     quant_preset: str = "balanced"
     cache_policy: str = "adaptive"
     rolling_window_messages: int = 24
-    llama_cpp_gpu_layers: int = 0
+    # 999 = every layer on the GPU (Metal). 0 = CPU mode. A fresh install must
+    # run on Metal: with 0 the server also drops flash attention, which the
+    # quantized KV cache requires, and fails to create a context at all.
+    llama_cpp_gpu_layers: int = 999
     llama_cpp_threads: int = -1   # -1 = auto-detect from CPU cores at startup
     llama_cpp_batch_size: int = -1  # -1 = auto-detect from platform at startup
     # Speed optimizations
@@ -365,7 +368,7 @@ def load_config() -> AppConfig:
         quant_preset=os.environ.get("LOCALCODE_QUANT_PRESET", runtime_data.get("quant_preset", "balanced")),
         cache_policy=os.environ.get("LOCALCODE_CACHE_POLICY", runtime_data.get("cache_policy", "adaptive")),
         rolling_window_messages=int(os.environ.get("LOCALCODE_ROLLING_WINDOW_MESSAGES", runtime_data.get("rolling_window_messages", 24))),
-        llama_cpp_gpu_layers=int(os.environ.get("LOCALCODE_LLAMA_CPP_GPU_LAYERS", runtime_data.get("llama_cpp_gpu_layers", 0))),
+        llama_cpp_gpu_layers=int(os.environ.get("LOCALCODE_LLAMA_CPP_GPU_LAYERS", runtime_data.get("llama_cpp_gpu_layers", 999))),
         llama_cpp_threads=int(os.environ.get("LOCALCODE_LLAMA_CPP_THREADS", runtime_data.get("llama_cpp_threads", 8))),
         llama_cpp_batch_size=int(os.environ.get("LOCALCODE_LLAMA_CPP_BATCH_SIZE", runtime_data.get("llama_cpp_batch_size", 128))),
         # These were saved but never loaded back (silent data loss on every

localcode-0.4.1/src/localcode/ui/FORK_COMMIT

@@ -0,0 +1 @@
+0992e70d60b424d6e091cc04d2ffb2bf3e2ed5a8

{localcode-0.4.0a1 → localcode-0.4.1}/src/localcode/ui/launch.py

@@ -25,8 +25,11 @@ WAIT_S = 240
 
 
 def _models_dir() -> Path:
+    """Where GGUFs live: LOCALCODE_MODEL_DIR (the documented override, also
+    honoured by config), LOCALCODE_MODELS_DIR (the older launcher name), else
+    the configured/default directory."""
     from localcode.models_catalog import model_dir
-    env = os.environ.get("LOCALCODE_MODELS_DIR")
+    env = os.environ.get("LOCALCODE_MODEL_DIR") or os.environ.get("LOCALCODE_MODELS_DIR")
     return Path(env).expanduser() if env else model_dir()
 
 
@@ -191,7 +194,23 @@ def main(model: str | None = None, project: str | None = None) -> int:
         if alias:
             argv += ["-m", f"localcode/{alias}"]
         frontend = subprocess.Popen(argv, cwd=str(project_dir), env=env)
-        return frontend.wait()
+        # Watch both: if the supervisor dies (killed, crashed), the UI would
+        # sit on "No model loaded" with a picker that never answers. End the
+        # session with a clear message instead.
+        while True:
+            rc = frontend.poll()
+            if rc is not None:
+                return rc
+            if sup.poll() is not None:
+                frontend.terminate()
+                try:
+                    frontend.wait(timeout=5)
+                except subprocess.TimeoutExpired:
+                    frontend.kill()
+                print(f"localcode: the model service stopped unexpectedly (see {rd / 'supervisor.log'}). "
+                      "Run `localcode` again.", file=sys.stderr)
+                return 1
+            time.sleep(0.5)
     finally:
         cleanup()
         sup_log.close()

{localcode-0.4.0a1 → localcode-0.4.1}/src/localcode/ui/server_cmd.py

@@ -44,7 +44,7 @@ def server_command(gguf: str, port: int, alias: str | None = None) -> list[str]:
     # Vision: same rule as classic localcode (runtime.py) — pass --mmproj when the
     # catalog says this model ships a projector and the sidecar is on disk.
     mm = mmproj_for(gguf)
-    if mm is not None:
+    if mm is not None and "--mmproj" not in cmd:
         cmd += ["--mmproj", str(mm)]
     return cmd
 

{localcode-0.4.0a1 → localcode-0.4.1}/src/localcode/ui/supervisor.py

@@ -671,7 +671,7 @@ def main() -> int:
     ap.add_argument("--port", type=int, required=True)
     ap.add_argument("--control-port", type=int, required=True)
     ap.add_argument("--server", required=True)
-    ap.add_argument("--models-dir", default=os.environ.get(
+    ap.add_argument("--models-dir", default=os.environ.get("LOCALCODE_MODEL_DIR") or os.environ.get(
         "LOCALCODE_MODELS_DIR", str(Path.home() / ".local/share/localcode/models")))
     ap.add_argument("--parent-pid", type=int, default=0, help="launcher PID; exit if it disappears")
     a = ap.parse_args()

{localcode-0.4.0a1 → localcode-0.4.1/src/localcode.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: localcode
-Version: 0.4.0a1
+Version: 0.4.1
 Summary: High-performance AI coding on consumer hardware.
 Author: LocalCode contributors
 License-Expression: Apache-2.0

{localcode-0.4.0a1 → localcode-0.4.1}/tests/test_comprehensive_bg_download.py

@@ -45,12 +45,19 @@ _CONTRACT_KEYS = {
 
 @pytest.fixture(autouse=True)
 def reset_registry():
-    """Wipe the shared background-download registry around each test."""
+    """Wipe the shared background-download registry around each test, and
+    wait for the worker threads a test started. A straggler from a previous
+    test (its gate released, thread not yet finished) otherwise marks the
+    next test's identically-keyed entry "done" mid-assertion."""
+    before = set(threading.enumerate())
     with bootstrap._DOWNLOAD_LOCK:
         bootstrap._DOWNLOADS.clear()
         bootstrap._DOWNLOAD_QUEUE.clear()
         bootstrap._DOWNLOAD_CHOICES.clear()
     yield
+    for t in threading.enumerate():
+        if t not in before and t is not threading.current_thread():
+            t.join(timeout=15)
     with bootstrap._DOWNLOAD_LOCK:
         bootstrap._DOWNLOADS.clear()
         bootstrap._DOWNLOAD_QUEUE.clear()
@@ -154,7 +161,9 @@ def test_concurrency_cap_leaves_extras_queued(monkeypatch, not_on_disk):
     release = threading.Event()
 
     def _gated_download(choice, on_progress=None, cancel_event=None):
-        release.wait(timeout=5.0)
+        # Hold until the test releases; a short timeout let a download "finish"
+        # before the queued assertion on slow CI runners.
+        release.wait(timeout=60.0)
         return True, str(choice.local_path)
 
     monkeypatch.setattr(bootstrap, "download_model", _gated_download)
@@ -177,8 +186,9 @@ def test_concurrency_cap_leaves_extras_queued(monkeypatch, not_on_disk):
     assert statuses.count("downloading") == 2
     assert statuses.count("queued") == 2
 
-    active = bootstrap.list_active_downloads()
-    assert len(active) == 4  # 2 downloading + 2 queued, terminal excluded
+    # 2 downloading + 2 queued, terminal excluded. The registry is updated
+    # from worker threads, so wait for the count rather than sampling once.
+    assert _wait_until(lambda: len(bootstrap.list_active_downloads()) == 4), bootstrap.list_active_downloads()
 
     # Free the slots; all four must drain to 'done'.
     release.set()

{localcode-0.4.0a1 → localcode-0.4.1}/tests/test_comprehensive_binary.py

@@ -121,10 +121,13 @@ def test_bundled_binary_loads_and_runs(binary: Path):
 
 
 def test_ui_binary_version_matches_package():
+    """The runtime footer shows the binary's version; it must be the same
+    release as the wheel (X.Y.Z). Pre-release suffixes may differ so an alpha
+    bump does not force a rebuild of the 98 MB binary."""
     path = BIN_DIR / "localcode-ui"
     if not path.is_file():
         pytest.skip("no bundled UI binary")
     from localcode import __version__
-    want = re.sub(r"(\d)(a|b|rc)(\d+)$", r"\1-\2\3", __version__)
+    want = re.match(r"\d+\.\d+\.\d+", __version__).group(0)
     out = subprocess.run([str(path), "--version"], capture_output=True, text=True, timeout=60, check=False).stdout.strip()
-    assert out == want, f"UI binary reports {out!r}, package is {want!r}; rebuild with scripts/build_ui_binary.sh"
+    assert out.startswith(want), f"UI binary reports {out!r}, package is {__version__!r}; rebuild with scripts/build_ui_binary.sh"

{localcode-0.4.0a1 → localcode-0.4.1}/tests/test_ui_launch.py

@@ -116,3 +116,4 @@ def test_supervisor_serves_every_route_the_ui_calls():
     src = (Path(entrypoint.__file__).parent / "ui" / "supervisor.py").read_text()
     missing = [r for r in CONTROL_ROUTES if f'"{r}"' not in src]
     assert not missing, missing
+

localcode-0.4.0a1/src/localcode/bin/localcode-ui

@@ -1,4 +0,0 @@
-[diffend] Oversized file quarantined before diffing.
-name: localcode-0.4.0a1/src/localcode/bin/localcode-ui
-size: 102794210 bytes
-sha256: af52ab40061bdad241ebc148d7d4c32a3140349c2f990c43b0e47511f3b8ff95

localcode-0.4.0a1/src/localcode/ui/FORK_COMMIT

@@ -1 +0,0 @@
-e0ca52b8bae67723ff15097b80a5531b6a74860a